Česká spořitelna finds customer segmentation is in the eye of the beholder

Though Česká spořitelna is the Czech Republic’s largest retail bank and has a long history, it discovered that it hardly knew its customers. A customer segmentation project with SAS software has given the bank the intelligence it needs to move forward with one-to-one marketing.

Česká spořitelna, with more than 5 million customers and roots that reach back to 1825, became a part of the powerful Erste Bank group in 2000. Since then, the company has undergone a significant transformation and improvement in key activities.

Česká spořitelna is now focused on improving its product offerings for individual clients. To do this, it has embarked on a project to segment its client base with greater sophistication, using data on the ownership and usage of the bank’s products and services, in addition to demographic data.

"By exploiting our customer data, we could base our segmentation on our own clients, not the entire market. This makes a big difference because we have a lot of information in our data warehouse and operational systems," says Petr Ptáčník, data mining project manager at Česká spořitelna.

Seeing customers as individuals
The bank’s business and marketing experts made it clear that precise client segmentation would be fundamental to future customer and product management. Above all, Česká spořitelna wanted to manage customers as individuals to maximize value.

An audit of the bank’s data suggested 10 dimensions around which to segment customers to give the bank the information it needed. The data mining team considered primary dimensions, which actually created the segment. Primary dimensions were subdivided into general client characteristics (profitability, life-cycle stage, "richness" and loyalty) and banking-specific characteristics (deposit versus loan clients, sophisticated banking clients versus traditional, active versus dormant, and clients with an active transactional account versus those without one).

Secondary dimensions, used not to create but rather to describe the segment and provide some additional insights into individual clients, included place of residence, further demographics and miscellaneous information about the client’s relationship with the bank. Several of the dimensions required some creative thinking and decision making to fill in the gaps in the information available to the team. For example, Česká spořitelna did not have figures on the individual profitability of customers, so the team decided to use gross earnings (interest plus non-interest) rather than true profitability.

The original plan was to use descriptive life-cycle stages such as child, young family, empty nest and so on, but the team found that the available data was often unreliable or incomplete. So the data miners had to make estimates based on available demographics such as age, income and outgoings, and student and pensioner indicators.

By "richness" the team meant estimation of financial potential: for example, what loans can the client afford, and what other earnings potential can the bank exploit? To calculate this value, the team added together the total credit amount on all accounts, half of the total debit amount and a function of net income. "We worked out a formula to calculate a client’s richness with considerable accuracy. But it was too complex a formula to be useful in sales and marketing, so we arrived at a simpler indicator," says Ptáčník.

Seven levels of loyalty
Česká spořitelna’s analysts were discovering that effective segmentation is as much an art as a science. Likewise, how do you decide if a customer is "loyal"? One aspect is the length of time that an individual has been the bank’s customer, of course. A second is the number of the bank’s products and services that the customer purchases, so the team integrated both into its evaluation of customer loyalty.

The team assigned seven levels of loyalty, ranging from non-clients in the firm’s databases through newcomers to "medium-bound clients" with two accounts, one of which must be more than five years old, up to "strong-bound clients" with at least three accounts.

"One of the most useful dimensions we found was the simple designation of deposit customers and loan customers, where we calculated the net balance by subtracting total loans from total deposits," says Ptáčník. "This was useful because it enabled us to identify customers with high deposits, whom we designated as ‘investors’ and those who are significantly in debt. Both of these groups are sources of value to the bank. Among the high-income customer segments, there tend to be very few customers who have zero or small balances, and a very large number of customers with a high positive or negative balance. These ratios are gradually reversed as you move down the income levels."

Simplifying specific segments
The marketers also wanted to be able to distinguish between traditional and more sophisticated clients, which sounds easy but turned out be more difficult in practice. "Usage of newer electronic distribution channels are indicative and useful as descriptors, but too narrow a definition," says Ptáčník. The "active versus dormant" clients was another interesting – and useful – dimension, based on the volume of transactions per month, which the team divided between "sleepers" at one extreme and "extremely active" at the other.

Modeling analysis simplified the assignment of clients to specific segments. "Again, the object of the exercise was to provide practical and useful information to sales and marketing," says Ptáčník. The team used standard statistical tools, such as correlation and factor analysis, and then used K-means to create 20 clusters using the 10 dimensions. "This can be quite complex, so we looked for ways to reduce the 20 clusters to 10 client segments (plus three non-client segments) by merging, re-segmenting and redesigning," says Jan Spousta, who performed the segmentation analysis. But if the aim was to come up with simple segments, why go to these lengths? Why not simply apply "off-the-shelf" rules for segmentation?

Spousta explains that this would not serve practical purposes, nor would it deliver the bank the competitive advantages it was seeking. For example, two of the client segments, depositors and subsidiary’s clients, are identical other than the fact that depositors are customers of the parent bank.

"The difference is meaningless from an analytical perspective. But we can only approach the depositors directly with offers, so from a subjective marketing perspective, it is of critical importance.

"The aim was to create some rules that would give us simple, hierarchical and easily updateable segments that will be of use to us in our specific applications. Segmentation is not just a statistical exercise. Segmentation is in the eye of the beholder," he adds.

Identifying the segments
The 10 client segments, in descending order of value to the bank, are:

• Investors: very high balance clients.
• Loan holders: loan and high-income clients, or mortgage holders.
• Solvent clients: high income or high balance.
• Debtors: significantly negative net balance.
• Youngsters and students: younger than 18 or student.
• Pensioners: older than 60 or pension is a primary income.
• Unsecured: low balance and owns a loan product.
• Transactors: average more than four transactions in a month.
• Subsidiary’s clients: not a customer of the parent bank.
• Depositors: others.

Value to the bank is the main determinant: for example, an investor is a high-balance client, regardless of other attributes such as age or income. "We believe that we have created a set of customer segments that the sales and marketing team is comfortable with," says Spousta. "In fact, a colleague in marketing actually asked, ‘Why is this so simple?’ which I regard as an indication of the correctness of our approach."

Finally, the segmentation project went into considerable detail in describing the segments: their size, profitability, demographics, product ownership and usage, and segment dynamics. For example, a chart plotting revenue by client on one axis and size of deposit on another showed investors to be of far higher value than any other segment – on the other hand, it is also the smallest segment. Loan holders, another very small segment, are also highly profitable, even though they have negative net balances. "Such observations give us important insights into how we should handle customers on an individual basis," says Spousta.

Expecting the unexpected
Česká spořitelna’s data miners made two segmentations of the customer base, one for 2002 and another for 2003, enabling them to track the transitions between segments. About 20 percent of clients moved between segments. The team mapped the most statistically significant movements between segments; many were expected, others were rather unexpected. "It helped us to understand typical life cycles, revealing the best ways to move customers from one segment up to a more profitable one. The typical route is from non-client to student or youngster, to depositor, transactor and (if the customer does well) to solvent client. Solvent clients then tend to move either to the investor segment or (for example, if the customer takes out a mortgage) to loan holder. Less successful customers tend to get stuck in a circular transition between the unsecured and debtors segments."

Česká spořitelna completed the description of the segments in the summer of 2004, supplementing it with the results of market research. "It is now being introduced into the life of our bank," says Spousta.

The data miners have already analyzed purchasing patterns within each of the segments, enabling the marketers to predict who will buy which products, and to focus marketing campaigns accordingly. "Predictive modeling based on segmentation will give us a key advantage in our market," concludes Ptáčník.

Copyright © SAS Institute Inc. All Rights Reserved.